Improving Crowded Object Detection via Copy-Paste

Crowdedness caused by overlapping among similar objects is a ubiquitous challenge in the field of 2D visual object detection. In this paper, we first underline two main effects crowdedness issue: 1) IoU-confidence correlation disturbances (ICD) and 2) confused de-duplication (CDD). Then explore pathway cracking these nuts from perspective data augmentation. Primarily, particular copy- paste scheme proposed towards making crowded scenes. Based on operation, design "consensus learning" method to further resist ICD problem then find out pasting process naturally reveals pseudo "depth" scene, which can be potentially used for alleviating CDD dilemma. Both methods are derived magical using copy-pasting without extra cost hand-labeling. Experiments show that our approach easily improve state-of-the-art detector typical detection task more than 2% any bells whistles. Moreover, work outperform existing augmentation strategies scenario.